International Journal of

Management Research Homepage: http://ijmres.pk/

Vol 10, No 2, 2020 June, PP. 215-226
and Emerging Sciences E-ISSN: 2313-7738, ISSN: 2223-5604

PORTFOLIO OPTIMIZATION WITH MEAN-VARIANCE & MEAN-CVAR:

EVIDENCE FROM PAKISTAN STOCK MARKET
1
Unbreen Arif, 2Dr. Muhammad Tayyab Sohail, 3Muhammad Irfan Majeed
1
National College of Business Administration & Economics. Email: unbreen@hotmail.com
2
National College of Business Administration & Economics. Email: tayyabsohail@yahoo.com
3
Bahria University Lahore Campus. Email: Irfan_imc@hotmail.com

ARTICLE INFO ABSTRACT

Article History:
Received: 10 Jan 2020
Revised: 21 Mar 2020
Accepted: 01 May 2020
Available Online: 30 Jun 2020
Keywords:
Portfolio optimization, Mean
CVaR, Mean Variance,
Pakistan Stock Market.
JEL Classification:
C61, G10, G11
The stock market in any country plays a significant role in economic
development through saving mobilization and providing diversification
opportunities. The investors and wealth managers always looked for those
securities, investment to which leads towards maximum returns. The
Markowitz portfolio optimization model is based on risk and reward trade off
where the returns are measured with Mean and Variance is used as proxy of
risk. The framework is criticized due to its normality of returns assumption and
in case of non-elliptical distribution the optimization results may lead towards
suboptimal portfolios. In emerging markets, returns are attributed to high
volatility, low liquidity and increased market concentration. Such volatile
phenomena signify the presence of downside risk. The expected shortfall or
CVaR has reported ability to measure loss beyond VaR and regarded as
coherent risk measure. This study investigated the impact on asset pricing of
PSX stocks with Mean Variance and Mean-CVaR model by analyzing KSE-
100 index stocks for the time period 2009-2018. The empirical findings of the
in sample performance of Mean-CVaR optimization for non-normal stocks
indicated higher Sharpe ratio while for normal stocks return data portfolio
selection and optimization with MV model SR is higher. The results of this
study are useful for the investors and fund managers to understand PSX
behavior and in selection of optimal portfolios.
© 2020 The authors, under a Creative Commons Attribution-Non-Commercial 4.0.

1. INTRODUCTION
Portfolio management and optimization are always of vital concerns for the investors and fund managers.
Markowitz (1952) regarded as pioneer who served the idea of portfolio optimization with Mean-Variance framework
according to investor’s sentiment of risk and reward. This work leads towards Modern Portfolio Theory (MPT), a
widely used tool for optimization of portfolios over the years. The goal of this normative theory is to establish a
balance between risk and reward. Under this normative approach the distribution returns are measured through mean
and its risk through Variance. This approach is simple and convenient for interpretations of results, but has been
criticized due to normality of return assumption as in real world normality of return is an ideal phenomenon. Moreover,
distribution explained with only two statistics might cause loss of vital information essential to construct efficient
portfolios. The Financial markets specifically in emerging economies are attributed to low liquidity, increased
volatility and asymmetric information (Snoussi & El‐Aroui, 2012), in presence of high tail risk modeling of asset
pricing is challenging . In Markowitz Mean Variance model risk is measured through only Variance since then other
risk measures introduced by many researchers, for example Value at Risk (VaR) is widely acceptable risk measure to
determine tail risk (Artzner P. , Delbaen, Eber, & Heath, 1999; Pflug, 2000; Uryasev, 2000) but criticized due to its
shortcomings i.e. its inability to satisfy subadditivity property, failure to capture severity of losses and its non-
convexity issue. To overcome the issues of VaR a new approach to minimize high losses risk, CVaR or Expected
shortfall was developed to controlled risk beyond VaR (Uryasev, 2000). Now a days Value-at-Risk (VaR) and
Conditional Value-at-Risk (CVaR) are widely popular risk management tool (Guo, Chan, Wong, & Zhu, 2018). But
CVaR is reported as consistent measure of risk and preferable over VaR (Rockafellar & Uryasev, 2000). CVaR also
known as tail VaR, mean shortfall or mean excess loss reported as coherent risk measure. In portfolio optimization
process CVaR used in two ways first as a constraint and aslo as minimization objective for portfolio choices (Uryasev,
2000).

215
Arif et al. Portfolio Optimization with Mean-Variance . . .

In Mean-Variance Optimization, Standard Deviation is used as risk measure which is a measure of total risk i.e.
measured both downside and upside risk but in real world investors are concerned about extreme losses not the gains.
Therefore in order to capture downside risk measure VaR was introduced with the evolution of financial markets and
financial regulatory body Basel Committee on Banking Supervision (BCBS) made VaR as supplement to credit risk
measure (Tian, Cai, & Fang, 2018). To enhance the process of portfolio selection and their optimization in presence
of non-elliptical distribution of returns and the efficiency of CVaR to measure downside risk there is a need to test the
portfolio optimization results obtained through Mean-CVaR and Mean-VaR in case of volatile distribution of returns.
The investment decisions are not merely dependent on reducing the Variance but investors also want to minimize the
downside risk. The research objective is attained by comparing optimization results of the portfolios formed under
Mean-Var, Mean-CVaR Model optimization. The portfolio selection and optimization results of Mean-CVaR are
compared with Mean-Variance portfolio choices to identify whether two optimization processes decide differently for
normal and non-normal stocks with ROI optimization method. ROI is used due to its reported efficiency to solve
optimization problem in consistent way for linear, quadratic , general non-linear and conic models (Theußl,
Schwendinger, & Hornik, 2020). This study contributed specifically to the literature of portfolio formation and
optimization process within the broad ambit of investment and portfolio Management.
The remainder of the paper is organized as follows: - The Section 2 incorporates brief literature review, the
Section 3 describes the research model data and methodology and the Section 4 presents the empirical findings and
results while section 5 consists of conclusions.
2. LITERATURE REVIEW
In the field of investment and portfolio management for construction of optimal portfolios the risk and reward
trade-off was first presented by Harry Markowitz (1952) which lead towards Modern Portfolio Theory. The Markowitz
(1952) Mean-Variance model remained widely acceptable tool, which addressed the investor’s sentiment for risk and
rewards. This approach is simple and convenient for interpretations of results while on the other side the distribution
explained with only two statistics might cause loss of vital information essential to construct efficient portfolios.
Moreover, Markowitz’s given normality of returns assumptions merely exist in real world as many studies have
reported the non-normality of returns (Kon, 1984; Fama E. , 1965; Harvey & A.Siddique, 2000). In emerging
economies financial markets returns are highly volatile and distributions of returns are non-normal (Luciano &
Marena, 2002; Asghar, Shah, Hamid, & Suleman, 2011; Snoussi & El‐Aroui, 2012; Sharif, 2018).
In presence of extreme events the returns might deviate from their average so the portfolio optimization with
mean-variance model can be misleading. This idea motivated many researchers to develop asset pricing model to
address the issue as in the presence of extreme events i.e. for Skewness and kurtosis the results obtained through MV
model are inappropriate as probability of extreme events are underestimated (P.Jones, 2010). For construction of
portfolio according to investors decision the higher moments are relevant (Samuelson, 1958 (Aracioğlu, Keskin, &
UCAK, 2011; Arif & Sohail, 2020). Many academic researchers supported for inclusion of higher moments in
construction of portfolios (Konno & Suzuki, 1992; Scott & P.A. Horvath, 1980; Lai, Yu, & Wang, 2006; K.Saranya
& Prasanna, 2014). The problem assocoiated with inclusion of higher moments are non convex, non smooth complex
optimisation. Morover, in practical world the investors are specifically concern with the downside volatility of returns
or loss and the jump or increase in returns might not be treated as risk. Unlike higher moments (skewness and excess
kurtosis), VaR has better ability to address the investors queries about “What is worst case scenario?” , “How much
amount of loss in bad times?” (Azher & Iqbal, 2018) as VaR estimate comprises of three factors “level of confidence,
time period & loss estimate”.
In Markowitz’s Mean Variance model, the risk is measured through Variance which has the ability to measure
risk on both sides or simply measure the deviation from the average values. Since then other risk measures have been
introduced by researchers (Artzner P. , Delbaen, Eber, & Heath, 1999; Uryasev, 2000; Pflug, 2000;). In 2007 ,Bali,
Gokcan, & Liang argued that VaR comprises of measureing ability of skewness, kurtosis and standard deviation and
reported it as better risk measure to capture market risk. The VaR is defined as the maximum loss over the given
confidence level (1-α). The concept of value at risk was introduced by Baumol (1963). In research reported that VaR
as an acceptable risk measure for financial institutes, fund managers and regulators. The VaR is regarded as easy and
practical approach to measure risk for portfolio managers (Miskolczi, 2016). In context of Pakistan Azher & Iqbal
(2018) conducted study where VaR is used as proxy to illiquidity factor in factor model as it subsumes Cokurtosis
explanatory power. Despite of the popularity of VaR it has been criticized due to its properties. It is reported by
(Artzner P. , Delbaen, Eber, & Heath, 1999) that VaR is coherent risk measure when it is based on SD of normal
distribution , further (McKey & Keefer, 1996) reported VaR inability to predict optimal position. Moreover, VaR’s
non-convexity and subadditivity property ignores fat tail risk (Luciano & Marena, 2002). To overcome such issues

216
 Int. J. Management Research & Emerging Sciences/10(2) 2020,215-226

CVaR or Expected Shortfall was introduced as efficient and coherent risk measure than VaR when the returns are
non-normalized (Uryasev, 2000). It is reported to be better performance measure than VaR (Uryasev, 2000; Artzner
P. , Delbaen, Eber, & Heath, 1999). CVaR being possessing superior mathematical properties has an ability to measure
and manage risk more efficiently (Uryasev, 2000). CVaR considered as better technique for portfolio optimization
and to evaluate risk (Rockafeller & Uryasev, 2000). CVaR has been reported as coherent risk measure (Pflug, 2000;
Ogryczak & Ruszczynski, 2002) and can be calculated through different techniques. It can be combined with analytical
or scenario-based methods to optimize portfolios with large numbers of instruments, in which case the calculations
often come down to linear programming. In order to address the inconsistencies in the mean variance frame work
(Roman, Dowman, & Mitra, 2007) merged Mean-CVaR in Mean Variance framework. A study conducted by
(Banihashemi & SarahNavidi, 2017) in context of Iran’s stock market revealed that the Mean-CVaR portfolio
performance is superior to Mean-VaR performance. Further as the confidence level increases the efficient frontier
move towards right wards and to get optimal results investor prefer to use higher confidence level. The empirical
results of the study conducted by Miskolczi (2016) by using 7 Hungarian stock market’s daily stocks data from 2005
to 2015 capture indicates that the Mean-variance optimization results significantly differ from Mean-CVaR portfolio
optimization results.
Pakistan is an emerging economy and its stock market showed a semi strong form of efficiency (shamshair, Baig,
& Mustafa, 2018) where the returns are highly volatile. The use of MPT will provide suboptimal result moreover most
of the studies in emerging economies lack in inclusion of CVaR in portfolio selection and optimization process. So
one of the objective of this study, to analyse portfolio selection and optimization with coherent risk measure and
compare with mean Variance frame work to identify whether the performance of the portfolio has been impacted
through the Man-CVaR Model over the traditional Mean Variance model.
3. DATA & METHODOLOGY
The Pakistan Stock Exchange is the only representative market for stock trading, it comes in existence in 2016
as three stock exchanges (Lahore, Karachi and Islamabad) merged together to single Stock Market. The KSE-100
index is used to measure performance of PSX and it was declared as Asia’s best in 2016. For the purpose of analysis
and valuation of portfolio selection and optimization of Mean-Var and Mean-CVaR model the KSE-100 index, 10-
years data is selected from 2009 to 2018. In Pakistan most of the mutual fund performance is reported on monthly
basis, moreover the monthly returns eliminate the element of noise. Therefore, monthly stock returns of 100 index
companies were analyzed who constantly remain part of index during study period. The total number of companies
which actively remained in KSE-100 index from 2009-2018 were 78. The monthly closing prices data of 78
companies’ stocks analyzed over the 10 years resulting total of 9,360 stock prices for the calculation of return and
analysis of portfolio optimization models. PSX is not efficient however a semi strong form of efficiency were observed
by many studies (Ali & Mustafa, 2001) (Nazir, Nawaz, Anwar, & Ahmed, 2010). The return KSE-100 index stocks
were calculated by substituting the price data in appended formula to calculate log returns, by following the
methodology of (Biglova, Jašić, Rachev, & Fabozzi, 2004)
𝑃𝑖𝑡
𝑅𝑒𝑡𝑢𝑟𝑛𝑠 = ln ( )
𝑃𝑖𝑡−1
Where 𝑃𝑖𝑡 is the current stock price & 𝑃𝑖𝑡−1 is the earlier / previous month stock price. The Data of monthly
stock prices was obtained from PSX and SBP websites. The portfolio selection and optimization model of Mean-Var
and Mean-CVaR models analyzed under certain constraints i.e. markets are perfect with no taxes and commission
costs, perfect division of assets by following standard assumptions of portfolio optimization of (Lai , 1991). Other
constraints are the maximum loss is equal to the total amount, absence of short sale and long only constraint. Further,
the time frame for portfolio selection and optimization is remain fixed from 2009 to 2018 of KSE-100 index.
To test the stock returns normality, we have used Jarque-Bera test as being identified the best for testing of
normality and the identification of stocks of non-normal returns. The rejection of null hypothesis is made on the basis
of p-value related to JB test. The acceptance of alternative hypothesis is determined by the value of p at 5% level of
significance. The normality test results of the KSE-100 index stocks are enclosed in Appendix.
Model Formulation for Portfolio Optimization Model:
Let 𝑤𝑛 is weight of n assets in portfolio, 𝑊 𝑡 is transpose vector of weights assign to portfolios. Let X is
distribution of stock returns the variables of the study are
Return / Mean = M = (𝑚1 , 𝑚2 , 𝑚3 , … … … . 𝑚𝑛 )𝑡 ,
V = Variance Covariance,
CVaR = Conditional Value at Risk

217
Arif et al. Portfolio Optimization with Mean-Variance . . .

𝑀𝑝 = E (X) = E [𝑊 𝑡 (𝑋 − 𝑋̅ ) ]
= ∑𝑛𝑖=1 𝑤𝑖 𝑚𝑖 …………………………………………………………….. (1)
𝑉𝑝 = V (X) = E [𝑊 𝑡 (𝑋 − 𝑋̅ )(𝑋 − 𝑋̅ ) ]
E [𝑊 𝑡 (𝑋 − 𝑋̅ )2 ]
∑𝑛𝑖=1 ∑𝑛𝑗=1 𝑤 𝑤𝑗 σ𝑖𝑗 Where 𝑖 ≠ 𝑗 …………………………………….(2)
𝑖

The variance co variance matrix of n x n assets is σ𝑖𝑗 , where variance is the diagonal of the matrix and other
observations are covariance of assets
σ11,𝑡 σ12,𝑡 … σ1𝑛,𝑡
σ21,𝑡 σ22,𝑡 … σ2𝑛,𝑡
[⋮ ⋮ ⋮ ]
σ𝑛1,𝑡 σ𝑛2,𝑡 … σ𝑛𝑛,𝑡

“Value at Risk or VaR is defined as the value of maximum that will not exceed from the given level of
confidence” (Richta'rik, 2015) . Let 𝑋 ′ random variable of loss then for given parameter 0<α<1 the 𝑉𝑎𝑅𝛼 for 𝑋 ′
is defined as:-

𝑉𝑎𝑅𝛼 = min {a: P (𝑋 ′ ≤ a) ≥ α} i.e. the minimum loss not exceeded with α………… (3)

Conditional value at risk (CVaR) also called expected tail loss or conditional tail loss or expected shortfall is
downside risk measure and can be calculated as;-
𝐶𝑉𝑎𝑅𝛼 = E [𝑋 ′ | 𝑋 ′ ≥𝑉𝑎𝑅𝛼 ] where 0<α<1

Discrete probability distribution for event 𝑌𝑗′ with probabilities 𝑃𝑗

1
𝐶𝑉𝑎𝑅𝛼 (𝑋 ′ ) = 1−𝛼 ∑𝑗:𝑓(𝑋 ′ ,𝑌 ′)≥𝑉𝑎𝑅 ′ 𝑃 𝑓(𝑋 ′ , 𝑌 ′ ) for (j=1,…..,n) …………….
𝑗 (4)
𝛼 (𝑋 )

And for continuous probability distribution, the calculation of ES or CVaR is as under:-

Let ϕ (.) be the standard normal cumulative distribution function and φ (.) is the standard normal density function
then for any confidence level t ϵ (0.5, 1)
ϕ(−𝑧𝑡) = (1 − 𝑡)
−𝑧𝑡
∫−∞ φ(x) 𝑑𝑥 = 1 − 𝑡
By using definition of VaR we can write it as
V [t,𝑥𝑤 ] = 𝑧𝑡 𝜎(𝑥𝑤 ) − 𝐸(𝑥)
CVaR is defined as the expected loss at confidence level by holding it over investment period where loss is ≥
VaR, therefore CVaR at 100% confidence level t is
Y [ t, 𝑥𝑤 ] = -E[𝑥𝑤 \ 𝑥𝑤 ≤ −V [t,𝑥𝑤 ]]
If the min –CVaR exist, then it will be mean variance efficient
Therefore, CVaR is
Y [t, 𝑥𝑤 ] = 𝑗𝑡 𝜎𝑥𝑤 − 𝐸(𝑥𝑤 )
−𝑧𝑡 𝑥 φ(x) 𝑑𝑥
Where 𝑗𝑡 = − ∫−∞ 𝑗𝑡 > 𝑧𝑡 & Y [t, 𝑥𝑤 ] > V [t,𝑥𝑤 ]
1−𝑡

In this study we have used R-studio for the calculation of Mean, Variance Covariance, VaR and CVaR and
optimization of portfolio has been made via ROI optimization method due to its reported efficiency to solve
optimization problem in consistent way for linear, quadratic , general non-linear and conic models. (Theußl,
Schwendinger, & Hornik, 2020).

218
 Int. J. Management Research & Emerging Sciences/10(2) 2020,215-226

4. EMPIRICAL RESULTS
We have performed the portfolio optimization for portfolio selection with an objective of Mean-Variance &
Mean-CVaR models through analysing stocks of KSE-100 index. First, we determined the normal and non-normal
stocks from the KSE-100 Index based on p-values. Out of 78 stocks 12 turns out to be normal. Approximately 84%
of the stocks are non–normal, signifying the presence of extreme events in determining returns of PSX. After
identifying the normal and non-normal stocks the portfolio selection and optimization is performed for Mean-Variance
and Mean-CVaR model. The optimization of the models of study is made under certain constraints i.e. markets are
perfect with no taxes and commission costs and perfect division of assets are available by following standard
assumptions of (Lai , 1991) for portfolio optimization. Another assumption is the maximum loss an investor can incurr
is about to the total amount, short sale is not allowed and the investor have long only constraint. Further assumptions
are time frame of stocks and portfolio allocations that is time period for optimal allocation is from 2009 to 2018 of
KSE-100 index stocks. Another constraint of this study is diversification of stocks through allocation of weight in
multiple stocks with limit i.e. 𝑤𝑖 ≥ 0. Moreover, in order to avoid concentration of more than 50% of funds in single
stock the portfolio allocation constraints are not more than 50% of funds in single stock and the minimum limit or
constraints for stock weight allocation is 0% to get diversification benefits.
The selection of portfolio and optimal weight allocation under MV and MCVaR model results can be seen in
Table-1.
Table 1. Asset allocation under Mean-Variance and Mean-CVaR optimization for non-normal stocks
MV Model Mean-CVaR Model
Stocks weights (%) Stocks weight (%)
HCAR 8% ABL 3%
LUCK 3% ICI 2%
MARI 9% INDU 7%
MLCF 3% JDWS 23%
PAKT 43% NESTLE 3%
PMPK 6% PAKT 22%
POML 7% POL 6%
SHFA 9% POML 13%
SML 12% SCBPL 3%
SHFA 14%
SML 4%
MEAN 0.03441 Mean 0.02536
SD 0.07914 SD 0.03931
Source: (Author’s estimation and calculation)

It is evident from the results of above table that the constructed portfolio under MV model is different from the
MCVaR model. In MV model the highest allocation of asset is made to PAKT i.e. 43% while in Mean-CVaR
optimization results indicated that the 45% allocation of funds is allocated to JDWS and PAKT respectively. Now
we plot the efficient frontier for both models and the efficient frontier provides the graphical representation of the
optimal portfolios. Figure 1 and 2 show the efficient frontier of MV and MCVaR portfolios where the returns are
allocated on the x-axis and the risk is scaled along y-axis. The concept of efficient frontier is attributed to the noble
laureate Markowitz (1952) work which is cornerstone of modern portfolio theory which provides the best optimal
portfolio solution with highest expected return for a given level of risk and the minimum risk for the given level of
return.

219
Arif et al. Portfolio Optimization with Mean-Variance . . .

Fig. 1. Mean-Variance Efficient Frontier for non-normal Fig. 2. Mean-CVaR Efficient Frontier for non-normal
stocks stocks
(Source: Author’s own estimation and calculation.)

The MV and Mean-CVaR efficient frontier plotted after obtaining Mean Variance optimization results of
monthly returns of non-normal stocks with monthly risk free rate of 0.65%. The portfolios which falls below this
frontier will give suboptimal results and the stocks which are above the portfolios are overpriced stocks. The two solid
dots on curve starting from lower indicate the minimum return and the upper solid black dot on curve indicate the
maximum expected return which can be obtained through optimal solution. It is clearly indicated from the MCVaR
efficient frontier that the optimal solution is obtained with minimal risk preference and Sharpe Ratio in Mean-CVaR
portfolio is 0.491 in comparison of Mean-Variance Sharpe ratio of 0.425 indicating the superiority of MCVaR in
sample performance for non-normal data. Now we implement the portfolio allocation and optimization model for
normal stocks of KSE-100 index. Table 2 presents the weighted allocation of stocks for both models.
Table 2. Asset allocation under Mean-Variance and Mean-CVaR optimization for normal stocks
MV Model Mean-CVaR Model
Stocks weights (%) Stocks weight (%)
FCCL 4.44% FCCL 8.68%
GADT 26.84% GADT 5.61%
MEBL 41.99% MEBL 50.80%
MTL 36.73% MTL 34.17%
MEAN 0.0363 MEAN 0.1179
SD 0.1614 ES 0.2239
(Source: Author’s own estimation and calculation)

The table 2 results clearly indicate that in case of normal returns the selection of stocks for portfolio in Mean-
Variance and Mean-CVaR model are same but different allocation of weights as for MV optimization model the
approximately of funds allocated to MEBL and MTL while in Mean-CVaR optimization model the 50% funds
allocation is made to MEBL followed it by 30% in MTL the stocks. Figure 3 & 4 show the efficient frontier of MV
and MCVaR model for normal stocks.

220
 Int. J. Management Research & Emerging Sciences/10(2) 2020,215-226

Fig. 3. Mean-Variance Efficient Frontier for normal stocks

Fig. 4. Mean-CVaR Efficient Frontier for normal stocks
(Source: Author’s own estimation and calculation)

The comparison of efficient frontier of figure 3 & 4 indicate that in case of normal distribution of returns the
optimal solution of Mean-Variance is above the maximum expected return while in Mean-CVaR optimization model
the optimal solution is below the maximum expected return which can be achieved through this model. Moreover, the
Sharpe ratio in MV optimization model is 0.22 in comparison of 0.153 Sharpe ratio of Mean-CVaR model, indicating
the in sample performance of MV model is higher in comparison of Mean-CVaR model for normal stocks return data.
In figure 5 the cumulative return performance of MV and Mean CVaR portfolio is presented for non-normal
distribution of returns .

Fig. 5. Cumulative Return Performance of MV & MCVaR model for time period 2009-2018

(Source: Author’s own estimation and calculation)

From 2015 onwards the cumulative return performance of Mean-Variance frame work is higher in comparison
with Mean-CVaR model due to the better performance of KSE-100 index in 2017 when the PSX was regarded as the
Asia’s best in terms of its performance. The empirical results indicate that during the recovery phase of PSX the
superiority of Markowitz MV model was proved and the results were aligned with the findings of (K.Saranya &
Prasanna, 2014).

221
Arif et al. Portfolio Optimization with Mean-Variance . . .

5. CONCLUSION
The results of current study clearly indicated that the portfolio selection choices and optimization under mean-
variance and Mean-CVaR model are impacted under normal and non-normal distribution of returns. Whereas in case
of non- normal distribution the Sharpe ratio of Mean-CVaR portfolio is higher in comparison of MV portfolio and
indicate the importance of Mean-CVaR portfolio optimization model for emerging economies financial markets
attributed to non- normal distribution of returns. The results of the model are prone to deviate with change of estimator
chosen for the purpose of calculation of VaR and CVaR. Moreover, the portfolio selection and formation can be
affected by the selection of length of time and holding period. The limitations of this study suggested future research
by analyzing portfolios of varying holding period and inclusion of other assets classes.
REFERENCES
Ali, S. S., & Mustafa, K. (2001). Testing Semi Strong Form Efficiency of Stock Market. The Pakistan Development
Review, 40(4), 651-674.
Aracioğlu, B., Keskin, F. D., & UCAK, H. (2011). Mean–Variance–Skewness–Kurtosis Approach to Portfolio
Optimization: An Application in Istanbul Stock Exchange. EGE AKADEMİK BAKIŞ / EGE ACADEMIC
REVIEW, 11(1), 9-17.
Arif, U., & Sohail, M. T. (2020). Asset Pricing with Higher Co-Moments & CVaR: Evidence from Paksitan Stock
Exchange. International Journal of Economics & Financial Issues, 10(5), 243-255.
Artzner, P., Delbaen, F., Eber, J., & Heath, D. (1999). Coherent measures of risk. Mathematical Finance, 9, 203-228.
Azher, S., & Iqbal, J. (2018). Testing Conditional Asset Pricing in Pakistan: The Role of Value-at-Risk and Illiquidity
Factors. Journal of Emerging Market Finance, 17(2S) 1S–23S. doi:DOI: 10.1177/0972652718777124
Bali, T. G., Gokcan, S., & Liang, B. (2007). Value at Risk and the Cross-Section of Hedge Fund Returns. JOurnal of
Banking and Finance, 31(4), 1135-1166.
Banihashemi, S., & Navidi, S. (2017). Portfolio performance evaluation in Mean-CVaR framework:A comparison
with non-parametric methods value at risk in Mean-VaR analysis. Operations research perspective, 4, 21-
28.
Banihashemi, S., & SarahNavidi. (2017). Portfolio performance evaluation in Mean-CVaR framework: A comparison
with non-parametric methods value at risk in Mean-VaR analysis. OPerations Research Perspective, 4, 21-
28.
Baumol, W. (1963). An Expected Gain-Confidence Limit Criterion for Portfolio Selection. Econ papers, 10(1), 174-
182.
Biglova, A., Jašić, T., Rachev, S., & Fabozzi, F. (2004). Profitability of momentum strategies:Application of novel
risk/return ratio stock selection criteria. Investment Management and Financial Innovations, 4, 48-62.
Fama, E. (1965). The Behavior of Stock Market Prices. Journal of Business, 38, 34-105.
Guo, X., Chan, R. H., Wong, W.-K., & Zhu, L. (2018). Mean-Variance, Mean-VaR, Mean-CVaR Models for Portfolio
Selection with Background Risk. SSRN electronic Journal. doi:https://dx.doi.org/10.2139/ssrn.3095099
Harvey, C., & A.Siddique. (2000). Condiional Skewness in Asset Pricing Tests. Journal of Finance, 1263-1295.
K.Saranya, & Prasanna, P. (2014). Portfolio Selection and Optimization with Higher Moments:Evidence from Indian
Stock Market. Asia Pacific Finance Markets.
Kon, S. (1984). Models of Stock Returns - a Comparison. Journal of Finance, 39, 147-165.
Levine, R., & Zervos, S. (1996). Stock Market Development and Long Run Growth. The World Bank Economic
Review, 10(2), 323-339.
Linsmeier, T., & Pearson, N. (2000). Value at risk. Financial Analysts Journal, 56, 47-67.
McKey, R., & Keefer, T. (1996). VaR is dangerous technique. Corporate finance searching for system integration
supplement.
Miskolczi, P. (2016). Differences Between Mean-Variance And Mean-Cvar Portfolio Optimization Models. Annals
of Faculty of Economics(1), 548-557.
Nazir, S. M., Nawaz, M., Anwar, W., & Ahmed, F. (2010). Determinents of stock Price volatility in Karachi Stock
Exchange: The mediating role of coporate dividend policy. international research Journal of Finance &
Economics, 55, 100-107.
Ogryczak, w., & Ruszczynski, A. (2002). Dual stochastic dominance and quantile risk measures. Int Trans Oper Res.
Pflug, G. C. (2000). Pflug GCh. Some remarks on the value-at-risk and the conditional valueat-risk. Probabilistic
Constrained Optimization: Methodology and Applications. In S. Uryasev (Ed.). Pflug GCh. Some remarks
on the value-at-risk and the conditional valueat-risk. Probabilistic Constrained Optimization: Methodology
and Applications.

222
 Int. J. Management Research & Emerging Sciences/10(2) 2020,215-226

Richta'rik, P. (2015). Optimization methods in finance lectcure notes .

Rockafellar, R. T., & Uryasev, S. (2000). Optimization of conditional value-at-risk. Journal of Risk, 2(3), 21-41.
doi:10.21314/JOR.2000.038
Rockafeller, R. T., & Uryasev, S. (2000). Optimization of conditional value-at-risk. Journal of Risk, 2(3), 21-42.
Roman, D., Dowman, K. D., & Mitra, G. (2007). Mean-risk models using two risk measures: a multi-objective
approach. Journal Quantitaive Finance, 7(4), 443-458.
Samuelson, P. (1958). “The fundamental approximation theorem of portfolio analysis in terms of means variances and
higher moments. Review of Economic Studies, 12, 65-86.
shamshair, M., Baig, M. j., & Mustafa, K. (2018). Evidence of random walk in Pakistan stock exchange: An emerging
stock market study. Journal of Economcs library, 5(1).
Snoussi, W., & El‐Aroui, M.‐A. (2012). Value‐at‐risk adjusted to the specificities of emerging markets: An analysis
for the Tunisian market. International Journal of Emerging Markets, 7(1), 86-100.
Theußl, S., Schwendinger, F., & Hornik, K. (2020). ROI:An Extensible ROptimization Infrastructure. Journal of
Statistical Software, 94(15).
Tian, D., Cai, Z., & Fang, Y. (2018). Econometric Modeling of Risk Measures: A Selective Review of the Recent
Literature. WORKING PAPERS SERIES IN THEORETICAL AND APPLIED ECONOMICS , 1-22.
Uryasev, S. (2000). Conditional Value at Risk: OPtimization algorithms and applications. Financial Engineering
News, 1-5.

223
Arif et al. Portfolio Optimization with Mean-Variance . . .

APPENDIX
Table Descriptive statistics and p-Value of KSE-100 Index Stocks

Stocks Mean St.Dev Variance Skewness Kurtosis CV p-value

ABL 0.0132490 0.0684599 0.0046868 1.0404605 5.088675 5.167172 5.72494E-10
ABOTT 0.0191971 0.082235 0.0067626 0.9303072 6.197667 4.283725 2.22045E-15
ACPL 0.0152765 0.0978478 0.0095742 0.5385335 4.118813 6.405097 0.002720356
AICL 0.0011868 0.0994497 0.0098902 0.244262 8.459326 83.794646 0
APL 0.0097686 0.0657109 0.0043179 0.3373877 5.682908 6.726753 5.91682E-09
ATLH 0.0116856 0.0937027 0.0087802 0.0125565 5.463582 8.018656 2.91516E-07
ATRL 0.0157917 0.114399 0.0130871 1.5904872 8.986616 7.24424 0
BAF 0.0124939 0.0696305 0.0048484 0.0564588 2.626702 5.573141 0.686402
BAHL 0.0095237 0.0581736 0.0033842 -0.1938788 3.520002 6.1083 0.3556675
BATA 0.0131956 0.1126281 0.0126851 1.3915141 6.391803 8.535252 0
BNWM 0.0081143 0.1351684 0.0182705 0.3552532 5.260212 16.657954 9.32837E-07
BOP 0.0083749 0.1188198 0.0141182 1.3211184 5.304217 14.187585 9.02611E-14
BWCL 0.0212822 0.1510597 0.022819 1.4929462 7.188005 7.097929 0
BYCO 0.0094793 0.1168321 0.0136497 1.3863993 5.990078 12.325016 0
CHCC 0.0200870 0.11266 0.0126923 0.8198316 4.436603 5.608606 9.03646E-06
CPPL 0.0213287 0.134509 0.0180927 1.1882017 5.690443 6.306495 1.89848E-14
DAWH 0.0075672 0.1208359 0.0146013 -1.1896743 11.902864 15.968329 0
DJKC 0.0186727 0.1011919 0.0102398 0.6488998 3.942549 5.419242 0.001886303
EFUG 0.0054123 0.0924185 0.0085412 1.1174289 8.157586 17.075729 0
ENGRO 0.0114870 0.0814464 0.0066335 0.3714696 3.635263 7.090286 0.09686022
FABL 0.0096511 0.0882839 0.007794 0.7654514 3.899505 9.147555 0.000466533
FCCL 0.0165151 0.0934915 0.0087407 0.168559 2.68277 5.660959 0.592052
FFBL 0.0101181 0.06663 0.0044396 0.2677728 3.259853 6.585237 0.4229188
GADT 0.0238194 0.1257337 0.015809 -0.0257125 3.84189 5.278628 0.1714316
GHGL 0.0030057 0.0863794 0.0074614 -0.1128659 6.051556 28.738194 8.32537E-11
GLAXO 0.0060694 0.0822851 0.0067708 1.6080962 9.959318 13.557465 0
HBL 0.008887 0.0794016 0.0063046 0.9794826 5.927871 8.934561 5.51781E-14
HCAR 0.0309188 0.1315333 0.017301 0.9053408 4.401018 4.254149 2.78847E-06
HMB 0.005458 0.0675408 0.0045618 0.3110622 3.128357 12.374605 0.3767184
HUBC 0.0161201 0.0583456 0.0034042 1.172332 6.795398 3.619428 0
HUMNL -0.0034459 0.1333675 0.0177869 -1.6213181 13.923737 -38.70292 0
IBFL 0.0096277 0.0979634 0.0095968 1.085166 6.094566 10.175141 5.55112E-16
ICI 0.0266152 0.105779 0.0111892 2.4295189 14.465737 3.974381 0
INDU 0.0266732 0.097731 0.0095514 1.3058721 12.203935 3.664013 0
INIL 0.014834 0.1004213 0.0100844 0.4868312 2.82543 6.76966 0.09378377
JDWS 0.020386 0.0867598 0.0075273 1.1259308 6.353972 4.255854 0
JGICL -0.000665 0.0756632 0.0057249 0.1888686 7.247467 -113.77258 0
JLICL 0.0267217 0.0870202 0.0075725 1.389514 10.543068 3.256539 0
JSCL 0.0052808 0.1703259 0.0290109 2.1750501 10.995278 32.253616 0

224
 Int. J. Management Research & Emerging Sciences/10(2) 2020,215-226

KAPCO 0.0039217 0.0497851 0.0024786 0.075372 4.170727 12.694758 0.03165467

KEL 0.0138202 0.1198242 0.0143578 1.8860013 7.853089 8.670214 0
KOHC 0.0252254 0.1333188 0.0177739 1.3210785 6.583232 5.285108 0
KTML 0.0269588 0.1416213 0.0200566 1.5097217 7.680061 5.253251 0
LUCK 0.0269386 0.086193 0.0074292 0.9051 6.716782 3.199607 0
MARI 0.0313516 0.1372763 0.0188448 1.0396643 7.429576 4.378611 0
MCB 0.0080548 0.0726276 0.0052748 1.1931486 5.456373 9.016649 3.36176E-13
MEBL 0.016474 0.0599336 0.003592 0.4227311 3.070468 3.638075 0.1755405
MLCF 0.0280955 0.140134 0.0196375 1.6813687 8.856107 4.987778 0
MTL 0.0176176 0.0778218 0.0060562 -0.2965133 3.481092 4.417265 0.2408071
MUREB 0.0259865 0.1302678 0.0169697 2.0461208 10.509455 5.012897 0
NATF 0.0176728 0.1244824 0.0154959 0.528806 7.328598 7.04374 0
NBP 0.0015016 0.0772281 0.0059642 -0.0510279 3.906418 51.431256 0.1271944
NCL 0.0203414 0.1069232 0.0114326 0.4757667 3.859941 5.256426 0.01792444
NESTLE 0.0205733 0.0850316 0.0072304 0.6769748 4.120329 4.133107 0.00053023
NRL 0.0121948 0.0942501 0.0088831 0.7454379 5.874556 7.728719 5.89173E-12
OGDC 0.0111942 0.0676015 0.00457 0.5698864 3.987516 6.038964 0.003859068
OLPL 0.0127581 0.0954599 0.0091126 0.1931386 4.936221 7.482282 6.41112E-05
PAEL 0.0130202 0.136915 0.0187457 0.7155574 3.530691 10.515571 0.003524119
PAKT 0.0366319 0.1216899 0.0148084 1.659154 6.150523 3.321969 0
PMPK 0.0280412 0.1347536 0.0181585 1.0759546 5.665124 4.805559 3.10418E-13
PICT 0.0211536 0.1131466 0.0128021 2.1148829 9.146733 5.348803 0
PIOC 0.0183629 0.1464737 0.0214545 1.0644758 7.385358 7.976627 0
PKGS 0.0142378 0.0864692 0.0074769 1.3043585 5.928126 6.073196 0
POL 0.0157865 0.0732112 0.0053599 0.3679388 4.045553 4.637583 0
POML 0.0225542 0.1131676 0.0128069 1.3848949 7.138461 5.017589 0
PPL 0.0033394 0.0596033 0.0035526 0.0407968 3.166014 17.848475 0.9190471
PSMC 0.0156965 0.0986564 0.0097331 0.6032087 3.635517 6.28526 0.01090135
PSO 0.0095799 0.0840597 0.007066 0.4643415 4.303332 8.774601 0.0018442
PTC -0.0004927 0.0748886 0.0056083 1.2566113 5.699875 -151.98558 3.33067E-15
SCBPL 0.0099828 0.0658683 0.0043386 0.8046955 4.137681 6.598157 7.72248E-05
SHEL 0.0049221 0.078997 0.0062405 0.2195452 3.5606 16.049509 0.2879114
SHFA 0.0241239 0.0977624 0.0095575 1.3596463 6.723966 4.052519 0
SML 0.0296827 0.1462997 0.0214036 1.3491014 5.40606 4.928779 1.33227E-14
SNBL 0.0037692 0.0905135 0.0081927 1.2498947 5.878234 24.014139 3.33067E-16
SNGP 0.0136899 0.1032841 0.0106676 0.2792946 2.69595 7.544573 0.3740753
SSGC 0.0089943 0.0951877 0.0090607 0.1735969 3.104158 10.58306 0.7274347
THALL 0.020339 0.1056596 0.011164 2.623275 18.430007 5.194939 0
TRG 0.0326925 0.1670759 0.0279144 1.395618 6.261428 5.110528 0
(Source: Author’s own estimation & calculation)

225
Arif et al. Portfolio Optimization with Mean-Variance . . .

Fig. 6. Optimal Weights under Mean-Variance Model for Fig. 7. Optimal Weights under Mean-CVaR Model for
normal stocks normal stocks

Figure 7 Optimal Weights under Mean-

Variance Model for non-normal stocks Figure 8 Optimal Weights under Mean-CVaR
Model for non-normal stocks

Fig. 8. Optimal Weights under Mean-Variance Model for Fig. 9. Optimal Weights under Mean-CVaR Model for
non-normal stocks non-normal stocks

(Source: Author’s own estimation & calculation)

226